Implementation of character based convolutional neural network

Last update: Nov 21, 2022

Overview

Character Based CNN

This repo contains a PyTorch implementation of a character-level convolutional neural network for text classification.

The model architecture comes from this paper: https://arxiv.org/pdf/1509.01626.pdf

There are two variants: a large and a small. You can switch between the two by changing the configuration file.

This architecture has 6 convolutional layers:

Layer	Large Feature	Small Feature	Kernel	Pool
1	1024	256	7	3
2	1024	256	7	3
3	1024	256	3	N/A
4	1024	256	3	N/A
5	1024	256	3	N/A
6	1024	256	3	3

and 2 fully connected layers:

Layer	Output Units Large	Output Units Small
7	2048	1024
8	2048	1024
9	Depends on the problem	Depends on the problem

Video tutorial

If you're interested in how character CNN work as well as in the demo of this project you can check my youtube video tutorial.

Why you should care about character level CNNs

They have very nice properties:

They are quite powerful in text classification (see paper's benchmark) even though they don't have any notion of semantics
You don't need to apply any text preprocessing (tokenization, lemmatization, stemming ...) while using them
They handle misspelled words and OOV (out-of-vocabulary) tokens
They are faster to train compared to recurrent neural networks
They are lightweight since they don't require storing a large word embedding matrix. Hence, you can deploy them in production easily

Training a sentiment classifier on french customer reviews

I have tested this model on a set of french labeled customer reviews (of over 3 millions rows). I reported the metrics in TensorboardX.

I got the following results

	F1 score	Accuracy
train	0.965	0.9366
test	0.945	0.915

Dependencies

numpy
pandas
sklearn
PyTorch 0.4.1
tensorboardX
Tensorflow (to be able to run TensorboardX)

Structure of the code

At the root of the project, you will have:

train.py: used for training a model
predict.py: used for the testing and inference
config.json: a configuration file for storing model parameters (number of filters, neurons)
src: a folder that contains:
- cnn_model.py: the actual CNN model (model initialization and forward method)
- data_loader.py: the script responsible of passing the data to the training after processing it
- utils.py: a set of utility functions for text preprocessing (url/hashtag/user_mention removal)

How to use the code

Training

The code currently works only on binary labels (0/1)

Launch train.py with the following arguments:

data_path: path of the data. Data should be in csv format with at least a column for text and a column for the label
validation_split: the ratio of validation data. default to 0.2
label_column: column name of the labels
text_column: column name of the texts
max_rows: the maximum number of rows to load from the dataset. (I mainly use this for testing to go faster)
chunksize: size of the chunks when loading the data using pandas. default to 500000
encoding: default to utf-8
steps: text preprocessing steps to include on the text like hashtag or url removal
group_labels: whether or not to group labels. Default to None.
use_sampler: whether or not to use a weighted sampler to overcome class imbalance
alphabet: default to abcdefghijklmnopqrstuvwxyz0123456789,;.!?:'"/\|_@#$%^&*~`+-=<>()[]{} (normally you should not modify it)
number_of_characters: default 70
extra_characters: additional characters that you'd add to the alphabet. For example uppercase letters or accented characters
max_length: the maximum length to fix for all the documents. default to 150 but should be adapted to your data
epochs: number of epochs
batch_size: batch size, default to 128.
optimizer: adam or sgd, default to sgd
learning_rate: default to 0.01
class_weights: whether or not to use class weights in the cross entropy loss
focal_loss: whether or not to use the focal loss
gamma: gamma parameter of the focal loss. default to 2
alpha: alpha parameter of the focal loss. default to 0.25
schedule: number of epochs by which the learning rate decreases by half (learning rate scheduling works only for sgd), default to 3. set it to 0 to disable it
patience: maximum number of epochs to wait without improvement of the validation loss, default to 3
early_stopping: to choose whether or not to early stop the training. default to 0. set to 1 to enable it.
checkpoint: to choose to save the model on disk or not. default to 1, set to 0 to disable model checkpoint
workers: number of workers in PyTorch DataLoader, default to 1
log_path: path of tensorboard log file
output: path of the folder where models are saved
model_name: prefix name of saved models

Example usage:

python train.py --data_path=/data/tweets.csv --max_rows=200000

Plotting results to TensorboardX

Run this command at the root of the project:

tensorboard --logdir=./logs/ --port=6006

Then go to: http://localhost:6006 (or whatever host you're using)

Prediction

Launch predict.py with the following arguments:

model: path of the pre-trained model
text: input text
steps: list of preprocessing steps, default to lower
alphabet: default to 'abcdefghijklmnopqrstuvwxyz0123456789-,;.!?:'"\/|_@#$%^&*~`+-=<>()[]{}\n'
number_of_characters: default to 70
extra_characters: additional characters that you'd add to the alphabet. For example uppercase letters or accented characters
max_length: the maximum length to fix for all the documents. default to 150 but should be adapted to your data

Example usage:

python predict.py ./models/pretrained_model.pth --text="I love pizza !" --max_length=150

Download pretrained models

Sentiment analysis model on French customer reviews (3M documents): download link

When using it:
- set max_length to 300
- use extra_characters="éàèùâêîôûçëïü" (accented letters)

Contributions - PR are welcome:

Here's a non-exhaustive list of potential future features to add:

Adapt the loss for multi-class classification
Log training and validation metrics for each epoch to a text file
Provide notebook tutorials

License

This project is licensed under the MIT License

Comments

Model trained on GPU is unable to predict on CPU

I used some GPUs on the server to speed up training. But after downloading the trained model file to my PC (no GPU equipped) and run the predict.py script. It gives an error message related to cuda_is_available() , seems that the model trained on a GPU cannot predict on only-CPU machines? Is this an expected behavior? If not, any help will be appreciated! Thanks a lot!

Error Message:

(ml) C:\Users\lzy71\MyProject\character-based-cnn>python predict.py --model=./model/testmodel.pth --text="I love the pizza" > msg.txt
C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py:454: SourceChangeWarning: source code of class 'torch.nn.modules.container.ModuleList' has changed. you can retrieve the original source code by accessing the object's source attribute or set `torch.nn.Module.dump_patches = True` and use the patch tool to revert the changes.
  warnings.warn(msg, SourceChangeWarning)
C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py:454: SourceChangeWarning: source code of class 'torch.nn.modules.container.Sequential' has changed. you can retrieve the original source code by accessing the object's source attribute or set `torch.nn.Module.dump_patches = True` and use the patch tool to revert the changes.
  warnings.warn(msg, SourceChangeWarning)
C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py:454: SourceChangeWarning: source code of class 'torch.nn.modules.conv.Conv1d' has changed. you can retrieve the original source code by accessing the object's source attribute or set `torch.nn.Module.dump_patches = True` and use the patch tool to revert the changes.
  warnings.warn(msg, SourceChangeWarning)
Traceback (most recent call last):
  File "predict.py", line 39, in <module>
    prediction = predict(args)
  File "predict.py", line 10, in predict
    model = torch.load(args.model)
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 387, in load
    return _load(f, map_location, pickle_module, **pickle_load_args)
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 574, in _load
    result = unpickler.load()
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 537, in persistent_load
    deserialized_objects[root_key] = restore_location(obj, location)
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 119, in default_restore_location
    result = fn(storage, location)
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 95, in _cuda_deserialize
    device = validate_cuda_device(location)
  File "C:\Users\lzy71\Anaconda3\envs\ml\lib\site-packages\torch\serialization.py", line 79, in validate_cuda_device
    raise RuntimeError('Attempting to deserialize object on a CUDA '
RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location='cpu' to map your storages to the CPU.

opened by desmondlzy 2

AttributeError: 'tuple' object has no attribute 'size'

train is always falling even with such kind of file: """ SentimentText;Sentiment aaa;1 bbb;2 ccc;3 """ Params of running -- just data_path Packages installed: numpy==1.16.1 pandas==0.24.1 Pillow==5.4.1 protobuf==3.6.1 python-dateutil==2.8.0 pytz==2018.9 scikit-learn==0.20.2 scipy==1.2.1 six==1.12.0 sklearn==0.0 tensorboardX==1.6 torch==1.0.1.post2 torchvision==0.2.1 tqdm==4.31.1

opened by 40min 2
Predict error

Raw output on console.

python3 predict.py --model=./models/model__epoch_9_maxlen_150_lr_0.00125_loss_0.6931_acc_0.5005_f1_0.4944.pth --text="thisisatest_______" --alphabet=abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_ Traceback (most recent call last): File "/Users/ttran/Desktop/development/python/character-based-cnn/predict.py", line 48, in <module> prediction = predict(args) File "/Users/ttran/Desktop/development/python/character-based-cnn/predict.py", line 11, in predict model = CharacterLevelCNN(args, args.number_of_classes) File "/Users/ttran/Desktop/development/python/character-based-cnn/src/model.py", line 12, in __init__ self.dropout_input = nn.Dropout2d(args.dropout_input) AttributeError: 'Namespace' object has no attribute 'dropout_input'

What is --number_of_classes argument? I don't have that set in the run command.

opened by thyngontran 1
Data types of columns in the data (CSV)

Can you describe how to encode the labels? I get only 1 class label, see output below. They are set as integers (either 0 or 1)

See output below when I train my model.

data loaded successfully with 9826 rows and 1 labels Distribution of the classes Counter({0: 9826})

opened by rkmatousek 1

RuntimeError: expected scalar type Long but found Double

I'm using a dataset I scraped but same structure comments with rating 0-10, using the same commands as provided except group_labels=0

Traceback (most recent call last):
  File "train.py", line 415, in <module>
    run(args)
  File "train.py", line 297, in run
    training_loss, training_accuracy, train_f1 = train(model,
  File "train.py", line 50, in train
    loss = criterion(predictions, labels)
  File "C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 532, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\loss.py", line 915, in forward
    return F.cross_entropy(input, target, weight=self.weight,
  File "C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\functional.py", line 2021, in cross_entropy
    return nll_loss(log_softmax(input, 1), target, weight, None, ignore_index, None, reduction)
  File "C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\functional.py", line 1838, in nll_loss
    ret = torch._C._nn.nll_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
RuntimeError: expected scalar type Long but found Double

opened by RyanMills19 0

Data loader class issues while mapping

I am using my dataset having three labels 0,1,2. While loading the dataset in data_loader class it generates key error. I think the issue is of mapping please guide.

Traceback (most recent call last):
  File "train.py", line 415, in <module>
    run(args)
  File "train.py", line 219, in run
    texts, labels, number_of_classes, sample_weights = load_data(args)
  File "/content/character-based-cnn/src/data_loader.py", line 55, in load_data
    map(lambda l: {1: 0, 2: 0, 4: 1, 5: 1, 7: 2, 8: 2}[l], labels))
  File "/content/character-based-cnn/src/data_loader.py", line 55, in <lambda>
    map(lambda l: {1: 0, 2: 0, 4: 1, 5: 1, 7: 2, 8: 2}[l], labels))
KeyError: '1'

opened by bilalbaloch1 1

ImportError: No module named cnn_model

Ubuntu 18.04.3 LTS Python 3.6.9

Command: python3 predict.py --model "./models/pretrained_model.pth" --text "I love pizza !" --max_length 150

Output: Traceback (most recent call last): File "predict.py", line 47, in prediction = predict(args) File "predict.py", line 14, in predict state = torch.load(args.model) File "/home/reda/.local/lib/python3.6/site-packages/torch/serialization.py", line 426, in load return _load(f, map_location, pickle_module, **pickle_load_args) File "/home/reda/.local/lib/python3.6/site-packages/torch/serialization.py", line 613, in _load result = unpickler.load() ModuleNotFoundError: No module named 'src.cnn_model'

opened by redaaa99 0

Releases(model_en_tp_amazon)

model_en_tp_amazon(Nov 17, 2019)

max length 1014 data : amazon + trustpilot number of classes : 3
Source code(tar.gz)
Source code(zip)
model_tp_amazon_1014.pth(43.25 MB)
english(Nov 6, 2019)

model trained on english customer reviews max_length = 500 no extra characters
Source code(tar.gz)
Source code(zip)
model_en.pth(23.25 MB)
trustpilot(Oct 30, 2019)
Model trained on french customer reviews

max length: 500

extra characters : "éàèùâêîôûçëïü"

Source code(tar.gz)
Source code(zip)
model.pth(23.33 MB)

Owner

Ahmed BESBES

Data Scientist, Deep learning practitioner, Blogger, Obsessed with neat design and automation

GitHub Repository

YOLOX-CondInst - Implement CondInst which is a instances segmentation method on YOLOX

YOLOX CondInst -- YOLOX 实例分割前言本项目是自己学习实例分割时，复现的代码. 通过自己编程，让自己对实例分割有更进一步的了解。若想

16 Nov 18, 2022

VIMPAC: Video Pre-Training via Masked Token Prediction and Contrastive Learning

This is a release of our VIMPAC paper to illustrate the implementations. The pretrained checkpoints and scripts will be soon open-sourced in HuggingFace transformers.

74 Dec 03, 2022

Source-to-Source Debuggable Derivatives in Pure Python

Tangent Tangent is a new, free, and open-source Python library for automatic differentiation. Existing libraries implement automatic differentiation b

2.2k Jan 01, 2023

A practical ML pipeline for data labeling with experiment tracking using DVC.

Auto Label Pipeline A practical ML pipeline for data labeling with experiment tracking using DVC Goals: Demonstrate reproducible ML Use DVC to build a

4 Mar 08, 2022

Machine learning Bot detection technique, based on United States election dataset

Machine learning Bot detection technique, based on United States election dataset (2020). Current github repo provides implementation described in pap

4 Nov 20, 2022

Code for the paper "Benchmarking and Analyzing Point Cloud Classification under Corruptions"

ModelNet-C Code for the paper "Benchmarking and Analyzing Point Cloud Classification under Corruptions". For the latest updates, see: sites.google.com

45 Dec 28, 2022

Code for our WACV 2022 paper "Hyper-Convolution Networks for Biomedical Image Segmentation"

Hyper-Convolution Networks for Biomedical Image Segmentation Code for our WACV 2022 paper "Hyper-Convolution Networks for Biomedical Image Segmentatio

17 Nov 02, 2022

Consistency Regularization for Adversarial Robustness

Consistency Regularization for Adversarial Robustness Official PyTorch implementation of Consistency Regularization for Adversarial Robustness by Jiho

40 Dec 17, 2022

Learning where to learn - Gradient sparsity in meta and continual learning

Learning where to learn - Gradient sparsity in meta and continual learning In this paper, we investigate gradient sparsity found by MAML in various co

28 Dec 09, 2022

Code for Multinomial Diffusion

Code for Multinomial Diffusion Abstract Generative flows and diffusion models have been predominantly trained on ordinal data, for example natural ima

104 Jan 04, 2023

Multistream CNN for Robust Acoustic Modeling

Multistream Convolutional Neural Network (CNN) A multistream CNN is a novel neural network architecture for robust acoustic modeling in speech recogni

37 Sep 21, 2022

[CVPR'21 Oral] Seeing Out of tHe bOx: End-to-End Pre-training for Vision-Language Representation Learning

Seeing Out of tHe bOx: End-to-End Pre-training for Vision-Language Representation Learning [CVPR'21, Oral] By Zhicheng Huang*, Zhaoyang Zeng*, Yupan H

196 Dec 13, 2022

Calculates JMA (Japan Meteorological Agency) seismic intensity (shindo) scale from acceleration data recorded in NumPy array

shindo.py Calculates JMA (Japan Meteorological Agency) seismic intensity (shindo) scale from acceleration data stored in NumPy array Introduction Japa

3 Sep 23, 2022